Just ten years after A. M. Bell introduced his vowel model in 1867, Henry Sweet (1877, A Handbook of Phonetics, Oxford, p. 18) wrote “After a time the student will be able to recognise each vowel solely by the muscular sensations associated with its formation”. This is the earliest instance I have found so far about this special faculty that phoneticians were expected to possess and practise. Sweet was very explicit: every vowel, and it was easily learnt. Needless to say, the “formation” of vowels referred to the articulations specified in the Bell model that Sweet was promoting with enormous success.
Daniel Jones (1932, Outline of English Phonetics, 3rd revised edition, §127) was slightly more cautious: “The positions of the tongue in the formation of the different vowel-sounds may, to a large extent, be felt”. However he did not explain which positions could be felt although “to a large extent” should mean a majority. Sadly, the tongue positions of the Bell model were only assumptions, and were ultimately discredited, so there was hardly anything there to be felt. Eventually, as the 20th century progressed, Sweet’s “muscular sensations” were linked to the neurophysiological terms kinesthesia and proprioception, although that is not how proprioception works (see any comprehensive handbook of neurophysiology, for example Siegel & Sapru, 2015, Essential Neuroscience, Wolters Kluwer, p. 253). Some aspects of proprioception are conscious, allowing awareness of some articulator positions, while others are nonconscious, preventing awareness of other articulator positions.
In conscious proprioception, receptors in joint capsules provide sensory information to the cerebral cortex where awareness of kinesthesia is generated. Only the mandibular joints would be relevant for speech articulation, and we are indeed good at feeling jaw movement. This had been recorded since the time of Panini, and jaw opening had always been a parameter of the ancient throat-tongue-lip model (Wood, 1993, The throat-tongue-lip model of vowel articulation, Phonum 2: 139-149, Dept. of Phonetics, Umeå university). Nonconscious proprioception, however, arises from muscle spindles and Golgi tendon organs and is relayed to sub-cortical motor centres such as the cerebellum solely for fine internal control of movement. For example, the spindles of the tongue muscles work silently, helping guide tongue manoeuvres for vowels. Consequently, we are not aware of this particular background activity working during vowel articulation. Feeling tongue positions for vowels by “muscular sensations” was an impossible feat.
What we do have is excellent conscious tactile sensation in the surface of the tongue that has always allowed good description of consonant articulations, and lingual contact with the maxillary molars and premolars during front vowels. The consequence of all this is that only gross tongue movements were reported for the throat-tongue-lip model – into the throat for [a]-like vowels, and up to the hard palate for [i-ɛ]-like vowels. A velar tongue posture for [u]-like vowels was only rarely mentioned in antiquity, but must have been implicit by the middle ages, once the lip position of [u-o] had been combined with the palatal tongue position of [i-ɛ] to accommodate [y-ø], described for example by the 12th century First Grammarian in Iceland (text and translation by Haugen 1950, First Grammatical Thesis, Language, Language Monograph 25). That was the extent of tongue articulation for vowels in the throat-tongue-lip model, that was rapidly replaced by the Bell model in the closing decades of the 19th century. Had anyone suggested that tongue positions might be felt with precision, they would have managed better than that during the few thousand years of the ancient model. And remember that Bell himself had used the ancient model in his best-selling (1863, Principles of Speech and Vocal Physiology, new editions to at least 1916).
Whoever was responsible, phoneticians were persuaded they could feel tongue positions with millimetre precision. One who had been convinced was Martin Joos (1933, Narrow transcription of General American, Le Maître Phonétique 48:48-50), disillusioned by 1948 (Acoustic Phonetics, Language Monograph 23, supplement to Language 24, pp. 55-56).